DIANNT
WorkAI systemsAldricAI

AI systems / 2026

AldricAI

A local LLM agent that reads an ASCII survival world, plans aloud, acts through IPC, and remembers across sessions.

An agent built on a Cataclysm: Dark Days Ahead fork, with a local neural backend and a persistent model of the game world.

AldricAI terminal showing the agent alongside ASCII game state
Reasoning, state, and game output remain visible during the local agent loop.
Role
Local game-agent engineer
Timeframe
2026
System
AI systems
  • game agents
  • local LLM
  • C++
  • IPC
  • persistent memory
  • ASCII games
01

Overview

The system in context.

AldricAI plays Cataclysm: Dark Days Ahead through a custom control layer. It reads structured game state, forms a natural-language plan, executes actions, and narrates its reasoning for viewers following the run.

The neural backend runs qwen3 locally, while a custom IPC path connects the model loop to a C++ game fork. Persistent memory tracks visited places, inventory, and encountered characters so each session begins with world context rather than a blank prompt.

What shipped

  • Reads structured state from a public Cataclysm: DDA fork.
  • Plans and narrates actions with a locally hosted qwen3 model.
  • Accepts viewer direction while the game-agent loop is running.
  • Persists world, inventory, location, and character context across sessions.
02

Measured context

The numbers, with their meaning intact.

local model backend
qwen3

The preserved project account identifies qwen3 as the on-device neural runtime.

game fork
C++

Cataclysm: DDA supplies the native simulation and fork surface.

Evidence source ↗ (opens in a new tab)
control bridge
IPC

A custom inter-process layer connects agent decisions to game actions.

03

Product record

Screenshots from the product and project record. Open any frame for a closer view.

01Reasoning, state, and game output remain visible during the local agent loop.
02The public fork is the simulation surface beneath the agent control layer.
04

Build story

From signal to shipped system.

  1. 01 / Perceive

    Expose the simulation state

    The game fork turns an ASCII world into structured observations the agent can reason over.

  2. 02 / Act

    Bridge thought and control

    Local model plans travel through IPC into valid game actions while reasoning remains visible.

  3. 03 / Remember

    Let the run accumulate meaning

    Persistent world memory carries places, inventory, and relationships into the next session.

Project links

Open the live product, working demo, publication, or repository.

Related work

All work →